May 7th, 2024 Fake people = Big business

Today’s newsletter was brought to you by Mercury.

Deeper Learning

Hey folks! It’s Tuesday and you know what that means. Time to go deeper. In today’s digest, we’re covering last week’s top AI product, a copilot for devs that’s not for coding, “RAG”, and some recent AI products you might have missed.

Let’s get into it!


AI enters uncanny valley territory

Spotting what’s human-made versus what’s AI-made is getting harder. Look at Midjourney. Sure, the early model was impressive, but you could easily spot something that was made with it (hint: just look at the hands.) Since then, AI has made some huge leaps, and now we’re heading into what I call uncanny valley territory.

Take these expressive new AI avatars from Synthesia. The startup has been honing AI business avatars since it launched four years ago, using its own AI model, Express-1. The avatars are used by Synthesia customers, like Zoom and IHG, for anything from marketing to customer support.

With tech makers like Synthesia continuously trying to make everything AI more realistic, deepfakes are a bigger concern than ever. And bad actors have indeed used Synthesia in the past, but fortunately the company’s B2B focus has kept the startup mostly in the clear. Plus Synthesia now takes extra precautions, like putting restrictions on the type of content people can make with its tools.

But that “fake” B2B salesperson or support rep - investors are here for it. Synthesia became a unicorn last year after hitting a $1 billion valuation with a Series C round that was led by Accel and included Nvidia.

When you’re building in the fastest-moving industry in the world, you need banking* and financial software that can keep up. Mercury simplifies your finances by powering them from your bank account.

We use Mercury at Product Hunt and the superior UX has helped us spend less time wrangling banks and more time building. You can tackle banking tasks in a click, control spend, and leverage built-in AI to process bills faster. Mercury gives us more precision, control, and focus.

Unlock exclusive perks on tools for builders in AI like AWS, Google Cloud, GoCharlie, and more when you sign up. Deposit $10K in your first 90 days and get a $500 cash bonus.


👗 Deep Fake Gala: Everyone’s favorite show & tell for the rich and famous happened last night. I’m, of course, talking about the Met Gala. This year AI made a guest appearance. People all around the internet shared AI-generated photos of the event. One of Katy Perry was so good it even fooled her mom.

🍎 Apple and AI: Apple isn’t quite ready to unveil its AI strategy just yet. But that doesn’t mean the company is staying silent either. At today’s iPad event, Apple made AI a big focus by talking up the potential for the iPad to make a big impact on the AI scene.


AI copilots aren’t just for coding

There’s no shortage of AI coding copilots out there — both GitHub and Amazon just launched their own last week. But where are the AI tools for non-coding related tasks, like say, writing technical documents or engineering design?

Eraser AI is a copilot not for coding but for generating technical design assets by writing natural language prompts. Eraser outputs diagram code that you can save, edit, and share with your team. Say you’re building a cloud infrastructure and you want to visually explain your thought process to your team. Instead of spending hours designing a diagram, you can tell Eraser what you need and it will get to work generating colorful, accurate, and icon-studded diagrams to your taste. 

Eraser first launched in 2020, founded by Shin Kim who formerly worked as the Chief of Staff for tech investor Elad Gil. You can follow along with Eraser’s progress here.


What the heck is RAG?

Question submitted by @rajiv_ayyangar.

Last week, we explained why AI models hallucinate, and this week we’re covering a related question submitted by one of our own — “What’s the deal with RAG?” 

You might have seen the term floating around the AI industry lately. It stands for Retrieval Augmented Generation and it’s a process that AI engineers can use to optimize the output of large language models. As you now know, hallucinations ultimately result from shortfalls in the training of AI — i.e. despite all the data that models are trained on, it’s hard to pack everything in. So what if you didn’t need to pack everything in? What if, instead, a model could retrieve information to enhance how accurate its responses are?

With RAG, AI engineers can introduce external data into the process in various formats (think records in databases, document files in repositories, or APIs). We’ll skip over the nitty-gritty of how the external data is delivered for now, but the main point is that it’s converted into a library that the models can “understand” (in short, they have numerical representations that help them determine what’s relevant). So when a user makes a search, the model is no longer just stuck with whatever input the user provided. The system can also reference that library of information and integrate that info with a person’s initial query so that the LLM can deliver contextually appropriate responses. 

For example, let’s say you’re asking an AI bot “How much PTO do I have left?” With RAG, the system would do some calculations on the documents it has access to, retrieve your company’s policy docs, and any requests you’ve made for time off this year. Then it would augment your original query and deliver the query to the LLM for an answer.

Got a question for us? Add it here!


What we've learned in 3 days of Llama 3” The good, bad, and ugly.


For Makers

  • AnswerTime helps you conduct user research with the help of AI.

  • Touch AI helps salespeople send personalized emails to highly curated leads.

For Work

  • AFFiNE is an AI assistant that helps you write, draw, and present stuff better.

  • Flownote transcribes and summarizes meetings right from your phone.

For Developers

  • GitHub launched a new Copilot tool to help you code faster and better.

  • Amazon also launched a new Copilot aimed at using AI to improve code. 

Thanks for going deeper with us!

Here via forward? Subscribe here.

Have feedback?

Did you enjoy today's newsletter?

Login or Subscribe to participate in polls.

Join the conversation

or to participate.